EE60039: Probability and Random Processes for Signals

and Systems
Instructor: Prof. Ashish R. Hota

Logistics:
• Class Timing: Monday: 12 noon - 12:55pm; Tuesday: 10am - 11:55am,
• Venue: NC 244
• Instructor Email: [email protected]. Use EE60039 in Subject Line.
• Course Website: http://www.facweb.iitkgp.ac.in/∼ahota/prob.html
• MS Team Password: ps6tv5h. All materials will be uploaded on Teams.

Syllabus:
Module A: Introduction to Probability and Random Variables. 4
Weeks. Main Reference: Chapters 1-5 of Wasserman.
1. Probability Space. Independence. Conditional Probability. [Chapter 1 of
Hajek, Chapter 2-5 of Chan, Chapter 1 of Gallager]
2. Random Variables and Vectors. Discrete and Continuous Distributions.
[Chapter 1 of Hajek, Chapter 2-5 of Chan, Chapter 1 of Gallager]
3. Expectation, Moments, Characteristic Functions. [Chapter 1 of Hajek, Chap-
ter 2-5 of Chan, Chapter 1 of Gallager]
4. Inequalities and Bounds. [Chapter 1-2 of Hajek, Chapter 6 of Chan, Chapter
1 of Gallager]
5. Convergence of Random Variables. Law of Large Numbers, Central Limit
Theorem. [Chapter 2 of Hajek, Chapter 6 of Chan, Chapter 1 of Gallager]
Module B: Random Processes. 4 Weeks.
1. Definition, Discrete-time and Continuous-time Random Processes [Chapter
4 of Hajek, Chapter 10 of Chan]
2. Stationarity, Power Spectral Density, Second order Theory [Chapter 4, 8 of
Hajek, Chapter 10 of Chan]

1
 3. Gaussian Process [Chapter 3 of Hajek, Chapter 3 of Gallager]
4. Markov Chain, Classification of States, Limiting Distributions [Chapter 4 of
Gallager]
Module C: Basics of Bayesian Estimation. 4 Weeks.
1. Maximum Likelihood, Maximum Aposteriori, Mean Square and Linear Mean
Square Estimation [Chapter 5 of Hajek, Chapter 8 of Chan]
2. Conditional Expectation and Orthogonality [Chapter 3 of Hajek, Chapter 10
of Gallager]
3. Kalman Filters [Chapter 3 of Hajek]
4. Hidden Markov Models [Chapter 5 of Hajek]
Module D: Information, Entropy, and Divergence, 1 Week

Reference:
The subject will closely follow the treatment in the following texts.
1. Larry Wasserman, All of Statistics, Springer Texts in Statistics, 2004.
Available at: https://link.springer.com/book/10.1007/978-0-387-21736-9
2. Bruce Hajek, Random Process For Engineers,
Cambridge University Press, 2015. Available at:
https://hajek.ece.illinois.edu/Papers/randomprocJuly14.pdf
3. Robert G. Gallager, Stochastic Processes: Theory for Applications,
Cambridge University Press, 2013.
4. Stanley H. Chan, Introduction to Probability for
Data Science, Michigan Publishing, 2021. Available at:
https://probability4datascience.com/index.html
5. Jason Speyer and Walter Chung, Stochastic Processes, Estimation
and Control, SIAM, 2008.

Evaluation Plan:
1. Midsem: 30%
2. Endsem: 50%
3. Homework and Class Performance: 20%
 Probability Space

Notations:
• N : set of natural numbers
• R : set of real numbers
• Z: set of integers
• Q : set of rational numbers
• R+ = {x ∈ R | x ≥ 0} and Z+ = {a ∈ Z | a ≥ 0}.
• For a set X, we denote the set of all its subsets by P(X)

Definition 1. Probability spaces are triplets of (Ω, F, P(·)) consisting

• Sample space: A set Ω that contains all possible outcomes.
• Events F: This is a set consisting of subsets of Ω satisfying:
a. Ω ∈ F,
b. Closed under complement: E ∈ F implies E c ∈ F, and
c. Closed under countable union: for any countably many sub-
sets E1 , . . . , Ek , . . . ∈ F, we have ∪∞
k=1 Ek ∈ F.

• Probability measure P(·): is a function from F to [0, 1] that satisfies:

i. P(Ω) = 1, and
ii. For countably many subsets {Ek } in F that are mutually disjoint
(i.e., Ei ∩ Ej = ∅ for all i 6= j), we have
∞
[ ∞
X
P( Ek ) = P(Ek ).
k=1 k=1

Note
• Any F that satisfies the properties a,b, and c is called a σ-algebra over Ω,
and (Ω, F) is called a measurable space.

1
 Examples

• Toss of a coin: Ω = {H, T }

• Roll of a dice: Ω = {1, 2, 3, 4, 5, 6}
• Waiting time for the next bus: Ω = {t ≥ 0}
• Each event is a subset of Ω.
• Event is ”yes/no questions that can be answered after the experiment is
conducted and the outcome is known”
• Example of measurable space: Ω = {0, 1}, and F = {∅, {1}, {2}, {1, 2}}.
• Example of measurable space: In general, power set P(Ω) is a σ-algebra for
any Ω.
• Example of measurable space: {∅, Ω} is a σ-algebra for any Ω.
• What if Ω is uncountable such as Rn ? Fortunately, for Rn there exists a
σ-algebra that we can define meaningful measures (such as uniform) in Rn ,
namely the Borel σ-algebra.
• Probability measure P measures the size of a set (an event).
• Example: Does the following define a probability space?
Ω1 = {1, 2, 3}, F1 = {φ, Ω, {1}, {2, 3}}
P1 [φ] = 0.5, P1 [{1}] = 0.3
P1 [Ω] = 0.2, P1 [{2, 3}] = 0.9.
• Countable Union: Example: (i) Ai = −1, 1 − 1i , i = 1, 2 . . .
 

Specifically,
S∞ A1 = [−1, 0], A2 = [−1, 0.5], . . . , A10 = [−1, 0.9]
i=1 Ai = {a | a ∈ Ai for some finite i} = [−1, 1].

• Homework: T∞
Bn [0, 1 − n1 ), n=1 Bn = ?
∞
Cn = [0, 1 + n1 ),
T
n=1 Cn = ?

2
 Elementary Properties implied by probability axioms

Let A, B ∈ F. Then, the following properties are true.

1. A ⊆ B, then P(A) ≤ P(B).
2. P (Ac ) = 1 − P(A).
3. P(A ∪ B) = P(A) + P(B) − P(A ∩ B) ≤ P(A) + P(B).
S  P
N N
4. P i=1 Ai 6 i=1 P (Ai ) for every N , including N = ∞. (Union bound)

3
 Conditional Probability and Independence

Definition 2. Consider a probability space (Ω, F, P). Consider events A and

B with P(B) > 0. The conditional probability of A given B is
P(A ∩ B)
P(A | B) := .
P(B)

Definition 3. A countable collection of events {A1 , A2 , . . .} are said to be mu-

tually independent, if any finite collection from the above, {A01 , A02 . . . A0k }
satisfies
P (A01 ∩ A02 . . . ∩ A0k ) = P (A01 ) · P (A02 ) · · · P (A0k ) .

Notes:
• If A, B are independent, P(B) > 0, then
P(A ∩ B) = P(A) · P(B) ⇒ P(A | B) = P(A).
Knowledge that event B is true gives you no further information about
occurence of A.
• Suppose A and B are disjoint can they be independent?
No. Disjoint is the strongest form of dependence. Occurence of one event
rules out the occurence of the other.

4
 Baye’s Law

Proposition 1. Let Ω be the set of outcomes. Let {A1 , A2 , . . . , Ak } form a

partition of Ω, and let B be another event. Then,
• {A1 ∩ B, A2 ∩ B, . . . , Ak ∩ B} also form a partition of B.
• Law of Total Probability:
k
X k
X
P(B) = P (Ai ∩ B) = P (B | Ai ) P (Ai ) .
i=1 i=1

• Baye’s Law:
P (Ai ∩ B) P (B | Ai ) P(Ai )
P (Ai | B) = = Pk .
P(B) j=1 P (B | A j ) P (Aj )

Problem: Consider a disease that affects one out of every 1000 individuals.
There is a test that detects the disease with 99% accuracy, that is, it clas-
sifies a healthy individual as having the disease with 1% chance, and a sick
individual as healthy with 1% chance. Then,
1. What is the probability that a randomly chosen individual will test pos-
itive by the test?
2. Given that a person tests positive, what is the probability that he or
she has the disease?

Homework: Repeat the above when detection accuracy is 99.9%, 99.99% and
99.999%.

5
 Random Variable

Definition 4. Let (Ω, F, P(·)) be a probability space. The mapping X :

Ω → R is called a random variable if the pre-image of any interval
(−∞, a] belongs to F, i.e.

X −1 ((−∞, a]) ∈ F for every a ∈ R, (1)

where

X −1 (B) := {ω ∈ Ω | X(ω) ∈ B}.

Note: Functions that satisfy this property are called measurable functions. Mea-
surability is a property of the function X and the σ-algebra.

Example: Let Ω = {HH, T H, HT, T T }. Consider two σ-algebras defined on Ω.

• F1 = {φ, Ω, {HH}, {HT, T H, T T }}
• F2 = {φ, Ω, {T T }, {HH, HT, T H}}
Consider a function Y : Ω → R such that Y (HH) = 1, Y (T H) = 1, Y (HT ) = 1
and Y (T T ) = 0, i.e, Y = 1 when at least one win toss is head, and 0, otherwise.
• Is Y a random variable with respect to F1 ?
• Is Y a random variable with respect to F2 ?

A random variable is neither random, nor is it a variable. The

function X itself is deterministic. Randomness is due to uncertainty regard-
ing which outcome ω ∈ Ω is true. Once the outcome ω is determined, the
value X(ω) is also determined.

6
 (Important) Indicator Random Variable

Definition 5. (Indicator Function) For a set E ⊆ Ω, define the indicator

function of E as

1 if ω ∈ E,
1E (ω) =
0 if ω 6∈ E.

Show that 1E (ω) is a random-variable if and only if (iff) E ∈ F.

Let us determine the pre-images for different a ∈ R. We have three cases;

1. if a < 0, 1−1
E ((−∞, a]) =?,

2. if 0 ≤ a < 1, then 1−1

E ((−∞, a]) =?, and

3. if 1 ≤ a, then 1−1
E ((−∞, a]) =?.

What do we conclude from here?

7
 Probability Distribution of a Random Variable

Definition 6. Consider a probability space (Ω, F, P(·)) and a random

variable X : Ω → R defined on this space. The probability distribu-
tion function of X is a function FX : R → [0, 1] defined as

FX (α) := P({ω : X(ω) ≤ α}) := P(X −1 (−∞, α]) := Prob(X ≤ α).

Example: Let Ω = {H, T }, F = 2Ω , P{H} = P{T } = 21 . Consider two random

variables.
• Y1 (H) = 1, Y1 (T ) = 0.
• Y2 (H) = 0, Y2 (T ) = 1.
Find the distribution functions FY1 and FY2 .

Properties of Distribution Function:

1. FX is non-decreasing, i.e, if α1 , ≤ α2 , FX (α1 ) ≤ FX (α2 ).
2. limα→∞ FX (α) = 1, limα→−∞ FX (α) = 0.
3. FX is right continuous, i.e., FX (α) = lim→0+ FX (α + ).

FX is called the cumulative distribution function (CDF).

8
 Example

Let Ω = {1, 2, 3} and X : Ω → R such that X(1) = 0.5, X(2) = 0.7, X(3) =
0.7.
• Find the smallest σ-algebra on Ω such that X is a random variable.
• Let P({1}) = 0.3. Find the distribution FX .

9
 Random Vectors and Random Processes

• Random Vectors: Any mapping X : Ω → Rn with X(ω) =

(X1 (ω), X2 (ω), . . . , Xn (ω)) is called a random vector if Xi is a random
variable for all i = 1, . . . , n.
• Random Process: An infinitely indexed collection {Xα }α∈I of random vari-
ables on (Ω, F, P) is called a random process.
• If the index set I is a discrete set (usually I = Z+ ), the random process is
called a discrete-time random process. When I = R or I = R+ , the random
process is called a continuous-time random process.

More generally, a random variable X maps one probability space (Ω1 , F1 , P1 )

to another (Ω2 , F2 , P2 ) in a systematic manner such that
• X : Ω1 → Ω2 ,
• for any event E2 ∈ F2 , its pre-image {ω ∈ Ω1 |X(ω) ∈ E2 } ∈ F1 , and
• for any event E2 ∈ F2 , P2 (E2 ) := P1 ({ω ∈ Ω1 |X(ω) ∈ E2 }) is called
the induced measure.

For a real-valued random variable X : Ω → R, the corresponding σ-algebra

on R is called the Borel σ-algebra and the induced measure gives rise to
the distribution function.

10
 Discussion on Random Variables

• When |Ω| is finite, we can define the collection of events F = 2Ω .

• However, then |Ω| is (uncountably) infinite, there are several technical diffi-
culties that arise in defining F = 2Ω .
• When Ω = R, we use a specific σ-algebra as the set of events.

Definition 7. The Borel σ- algebra, denoted B(R), is the smallest σ- algebra

that contains all sets of the form (−∞, α] for every α ∈ R.

• Define F := {(−∞, α]|α ∈ R}. Is F a σ- algebra ?

• B(R) = σ{F} is the σ- algebra generated by the sets contained in F.
• Thus, a real-valued random variable X : Ω → R is a mapping from an
arbitrary probability space (Ω, F, P) to (R, B(R), PX ) where the induced
measure PX is characterized in terms of the distribution function FX .

11
 Discrete Random Variable

• A random variable is discrete when X takes a finite or countable number

of values.
• Suppose X takes values in the set {x1 , x2 , ....xn }. Then,

P(X = x1 ) = P({ω ∈ Ω|X(ω) = x1 }) =: pX (x1 ), . . .

P(X = xn ) = P({ω ∈ Ω|X(ω) = xn }) =: pX (xn ).

where pX is called the probability mass function.
Pn
• The quantities {pX (x1 )....pX (xn )} satisfy pX (xi ) ≥ 0 and i=1 pX (xi ) = 1.
• The distribution function FX is a stair case function.
• Example: Bernoulli Random variable (p):
X = 1 with probability p and X = 0 with probability 1 − p.
• Binomial r.v (n, p):
Outcome of n coin tosses where each coin toss comes Head with probability
p.
Ω = {HH...H...T, ...T HHT..., T T...T }
X : Ω → R gives the number of Heads in n coin tosses.
E.g., X(HH...H) = n, X(HT T...T ) = 1, and so on.
Can we express a Binomial r.v in terms of a collection of Bernoulli r.v.s?
• Homework: write a program to plot the pmf and distribution of a Binomial
r.v. for n = 25 and p ∈ {0, 0.2, 0.4, 0.6, 0.8, 1}.

12
 Continuous Random Variable

• A random variable X is continuous if there exists a function fX : R → [0, ∞]

such that for every α ∈ R, we have
Z α
FX (α) = P({ω ∈ Ω|X(ω) ≤ α}) = fX (x)dx.
−∞

• fX : is called the probability density function (pdf ) of r.v. X.

dFX (x)
• If FX is differentiable at α, fX (α) = dx |x=α .
• Example: X is uniformly distributed between [a, b]. The pdf is given by
(
1
b−a , when α ∈ [a, b],
fX (α) =
0, otherwise.
Determine the distribution function FX .
• Example: Let X be a r.v which takes value 0 with probability 0.5. Otherwise,
it is uniformly dist. between 0.5 to 1.
1. Plot FX (α) for α ∈ R.
Rα
2. Find fX (x) such that −∞ fX (x)dx = FX (α).
• Exponential r.v. (λ > 0): The pdf is given by The pdf is given by
(
λe−λα , when α ≥ 0,
fX (α) =
0, otherwise.
Determine the distribution function FX .
−(α−µ)2
• Gaussian r.v. (µ, σ > 0): The pdf fX (α) = √ 1 e( 2σ 2
)
for α ∈ R.
2πσ

13
 Properties of probability density function

For a continuous random variable X, its pdf satisfies the following properties.
1. fX (x) ≥ 0, for every x ∈ R.
R∞
2. −∞ fX (x)dx = FX (∞) = 1.
3. fX (x) is not a probability; if can be take values larger than 1 at some points.
R x+ Rx R x+
4. FX (x + ) − FX (x) = −∞ fX (x)dx − −∞ fX (x)dx = x fX (x)dx.
Rb
5. P(a ≤ X ≤ b) = FX (b) − FX (a) = a fX (x)dx.

Note: If the CDF of a r.v. X is continuous at some α, then P(X = α) = 0.

14
 Expectation of a Random Variable

• A r.v. X is called a simple random variable if it takes finite number of

possible values, i.e.,

 a1 ,
 if ω ∈ A1
X(ω) = a2 , if ω ∈ A2 , . . .

an , if ω ∈ An .


For this simple r.v X, we define E[X] := ni=1 ai P(Ai ) ∈ R.

P

• Indicator r.v for event A is a simple random variable with E[1A ] = P(A).

• For a non-negative random variable X,

– there exists a sequence of simple random variables {X1 , X2 , . . .} which
converges to X.
– the expectation of each simple r.v in the sequence {E[X1 ], E[X2 ], . . .}
can be computed as above, and this sequence of real number is conver-
gent,
– E[X] := limn→∞ E[Xn ].
• For discrete r.v: E[X] = ni=1 xi P(X = xi ) = ni=1 xi pX (xi ).
P P
R∞
• For continuous r.v: E[X] = −∞ xfX (x)dx
R R
• General notation: E[X] = xdFX (x) = Ω X(ω)dP(ω).

• Note: E[X] ∈ R, i.e., expectation of a random variable is a deterministic

scalar without any randomness in it.

15
 Properties of Expectation

Definition 8. For two random variables X and Y , we define

• X = Y almost surely (a.s.) if P [{ω ∈ Ω|X(ω) = Y (ω)}] = 1.
• X ≤ Y almost surely (a.s.) if P [{ω ∈ Ω|X(ω) ≤ Y (ω)}] = 1.

Properties of Expectation:
• Linearity: For two random variables X, and Y ,

E[αX + βY ] = αE[X] + βE[Y ] for any α, β ∈ R.

Equivalently, E[αX] = αE[X], & E[X + Y ] = E[X] + E[Y ].

• If X = Y a.s, then E[X] = E[Y ].

• If X ≤ Y a.s, then E[X] ≤ E[Y ].

16
 Function of random variables

• Let X be a random variable. Then Y = g(X) is a random variable if the

function g is measurable, i.e., for any α ∈ R, the inverse map Y −1 (−∞, α]
belongs to the Borel σ-algebra over R.

• All continuous functions are measurable. In fact, almost all functions we

encounter satisfies this property.

• Example: If X is a random variable, so are sin(X), log(X), X k , and so on.

• Law of the unconscious statistician (LOTUS): If Y = g(X), then

Z Z ∞
E[Y ] = ydFY (y) = g(x)dFX (x).
−∞

There is no need to find the distribution of Y . Thus, for a continuous r.v.

X with density fX , we have
Z ∞
2
E[X ] = x2 fX (x)dx
Z−∞∞
k
E[X ] = xk fX (x)dx
−∞
Z ∞
E[sin(X)] = sin(x)fX (x)dx.
−∞

• E[X k ] is called the k-th moment of X.

• Variance of a r.v. X is defined as E[(X − E[X])2 ] = E[X 2 ] − (E[X])2 .

17
 Characteristic Function

Characteristic function of a r.v X is defined as

√
CX (h) = E[eihX ], where i = −1.
R∞
• For a continuous r.v, CX (h) = −∞ eihX fX (x)dx.

• CX (0) = E[1] = 1.

dCX (h) R∞ ihx

• dh = −∞ (ix)e fX (x)dx.

dCX (h) R∞
• dh |h=0 = −∞ (ix)fX (x)dx = iE[X].

• How about higher order derivatives?

18
 Random Vector

 
X1
X 
 2
• A random vector X =  ..  such that each Xi , 1 ≤ i ≤ n is a r.v..
 . 
Xn

• Joint distribution function (CDF) FX : Rn → [0, 1] is defined as

FX (c1 , c2 , ....cn ) = P[{ω ∈ Ω|X1 (ω) ≤ c1 , X2 (ω) ≤ c2 , . . . , Xn (ω) ≤ cn }]
= P[∩ni=1 {ω ∈ Ω|Xi (ω) ≤ ci }].

• The random variables X1 , X2 , ....Xn are jointly continuous if there exists a

function fX : Rn → R≥0
Z c1 Z c2 Z cn
FX (c1 , c2 , . . . , cn ) = ... fX (x1 , x2 , ....xn )dx1 dx2 ....dxn .
−∞ −∞ −∞

• Random vector X = [X1 , . . . , Xn ]> is jointly discrete if each Xi is a joint

discrete random variable. Joint pmf is defined as
pX (c1 , c2 , ....cn ) = P({ω ∈ Ω|Xi (ω) = ci , 1 ≤ i ≤ n}).

• Joint Characteristic Function: For a continuous random vector X,

CX (h1 , h2 , ...hn ) = E[ei(h1 X1 +h2 X2 +....hn Xn ) ]
Z ∞Z ∞ Z ∞
= .... ei(h1 x1 +h2 x2 +...hn xn ) fX (x1 , x2 , ....xn )dx1 dx2 dxn .
−∞ −∞ −∞
 
E[X1 ]
 E[X ] 
2 
• Expectation: E[X] =  ..  ∈ Rn .

 . 
E[Xn ]

19
 Computing Marginal Distributions

If joint distribution/ density/ mass function is given, we can compute the distri-
bution/ density/ PMF of each individual constituent random variable.

• Joint distribution FX (c1 , c2 , . . . , cn ) = P[∩ni=1 {ω|Xi (ω) ≤ ci }].

• Marginal distribution of the second constituent random variable

FX2 (c2 ) = P[{ω ∈ Ω|X2 (ω) ≤ c2 }]

= P[∩ni=1,i6=2 {ω|Xi (ω) ≤ ∞} ∩ {ω ∈ Ω|X2 (ω) ≤ c2 }]
= lim lim . . . lim FX (c1 , c2 , . . . , cn )
c1 →∞ c3 →∞ cn →∞

• Suppose joint density fX (c1 , c2 , . . . , cn ) is given, Find fX2 (c2 ). Recall that
Z c1 Z c2 Z cn
FX (c1 , c2 , . . . , cn ) = ... fX (x1 , x2 , . . . , xn )dx1 dx2 . . . dxn
x1 =−∞ x2 =−∞ xn =−∞
FX2 (c2 ) = lim FX (c1 , c2 , . . . , cn )
ci →∞,i6=2
Z ∞ Z c2 Z ∞
= ... fX (x1 , x2 , . . . , xn )dx1 dx2 . . . dxn
Zx1c=−∞
2
x2 =−∞
Z ∞
xn =−∞
Z ∞ 
= ... fX (x1 . . . xn )dx1 dx3 . . . dxn dx2
Zx2 =−∞
c2
x1 =−∞ xn =−∞

=: fX2 (x2 )dx2

x2 =−∞

20
 Example

 
X
Consider a random vector with joint density
Y
(
x + cy 2 , x ∈ [0, 1], y ∈ [0, 1]
fXY (x, y) =
0, otherwise.

• Find the value of c.

• Find marginal densities fX (x) and fY (y).
• Find the cumulative distribution function FXY (c1 , c2 ).
• Compute P[0 ≤ X ≤ 21 , 0 ≤ Y ≤ 12 ] using the density and the cumulative
distribution function.

21
 Independence of Random Variables

A collection of random variables {X1 , X2 , . . . , Xn } are said to be mu-

tually independent if for any collection of Borel subsets (events on R)
{A1 , A2 , . . . , An } the underlying events {ω ∈ Ω|X1 (ω) ∈ A1 }, {ω ∈
Ω|X2 (ω) ∈ A2 } . . . are mutually independent.

We have the following equivalent conditions that are easier to verify.

• Joint CDF satisfies the following property.

FX (c1 , c2 , . . . , cn ) = P[∩ni=1 {ω|Xi (ω) ≤ ci }]

= Πni=1 P[{ω|Xi (ω) ≤ c1 }]
= FX1 (c1 ) × FX2 (c2 ) × . . . × FXn (cn )

• For a discrete set of random variables, independence is equivalent to joint

pmf satisfying

pX (c1 , . . . , cn ) = pX1 (c1 ) × . . . × pXn (cn ).

• For a continuous set of random variables, independence is equivalent to joint

pdf satisfying

fX (c1 , . . . , cn ) = fX1 (c1 ) × . . . × fXn (cn ).

• Joint characteristic function satisfies

CX (h1 , . . . , hn ) = CX1 (h1 ) × . . . × CXn (hn ) ∀{h1 h2 . . . hn }.

Only checking CX (h, h, . . . , h) = CX1 (h) × . . . × CXn (h) is not enough to

conclude. that Xi ’s are independent.
• E[X1 X2 . . . Xn ] = E[X1 ] × . . . × E[Xn ].
• More generally, for any collection of bounded continuous functions {g1 , g2 , . . . , gn },
E[g1 (X1 )g2 (X2 ) . . . gn (Xn )] = E[g1 (X1 )] × . . . × E[gn (Xn )].

22
 Practice Problems

Let X and Y have joint density

(
2e−(x+2y) , ifx > 0, y > 0,
fXY (x, y) =
0, otherwise.
Determine whether X and Y are independent.

Consider a random variable X with cumulative distribution function given

by: (
1 − 3−bxc , x ≥ 0,
FX (x) =
0, otherwise,
where bxc is the floor of x, i.e., the largest integer smaller than or equal to
x. Is X a discrete or continuous random variable? Compute P[X = 2] and
P[X > 2].

Let X and Y be two independent random variables, each having uniform

distribution over the range [0, 1]. Let Z = max(X, Y ) and W = min(X, Y ).
1. Determine the CDF and expectation of Z.
2. Determine the CDF and expectation of W .
3. Determine the covariance cov(Z, W ).

23
 Correlation and Covariance

Correlation between two random variables X and Y is defined as E[XY ].

Let X and Y be discrete random variables that take values as X ∈ {x1 , x2 .....xn }
and Y ∈ {y1 , y2 ......ym }. Let the joint pmf be pij = P(X = xi , Y = yj ). Then,
n X
X m
E[XY ] = xi yj pij = xT P y, where,
i=1 j=1
     
x1 y1 p11 p12 . . . p1m
x  y  p p . . . p 
 2  2  21 22 2m 
x =  ..  , y =  ..  , P = .. .
.  .   . 
xn ym pn1 pn2 . . . pnm

Covariance between two random variables X and Y is

 
cov(X, Y ) = E (X − E[X])(Y − E[Y ])
 
= E XY − XE[Y ] − Y E[X] + E[X]E[Y ]
= E[XY ] − E[X]E[Y ]

Correlation Coefficient
cov(X, Y )
ρX,Y = p p , −1 ≤ ρXY ≤ 1.
var(X) var(Y )
Inner product interpretation
n
T
X xT y
x y= xi yi = ||x|| ||y|| cos θ =⇒ cos θ = .
i=1
||x|| ||y||

If Y = aX + b, then determine ρX,Y .

24
 Properties of Covariance

Covariance satisfies the following properties.

• cov(X, X) = var(X).
• cov(X, Y ) = cov(Y, X).
• cov(aX, Y ) = acov(X, Y ).
• cov(X + c, Y ) = cov(X, Y ).
• cov(X + Z, Y ) = cov(X, Y ) + cov(Z, Y ).
• More generally,
Xn m
X n X
X m
cov( ai X i , bj Yj ) = bj ai cov(Xi , Yj ).
i=1 j=1 i=1 j=1

Two random variables X and Y are said to be uncorrelated if

cov(X, Y ) = 0. If X and Y are independent, then they are uncorrelated.
However, the converse is not true.

25
 Covariance Matrix of a Random Vector

 
X1
X 
 2
For a random vector X =  .. , the covariance matrix contains the covariance
 . 
Xn
of each pair of constituent random variables.
 
cov(X1 ) cov(X1 , X2 ) . . . cov(X1 , Xn )
cov(X, X) = cov(X) =  ..  ∈ Rn×n
.
cov(Xn , X1 ) cov(Xn , X2 ) . . . cov(Xn )
 
X1 − E[X1 ] (X1 − E[X1 ]) (X2 − E[X2 ]) . . . (Xn − E[Xn ])
 X − E[X ] 
 2 2
= E

.. 
 . 
Xn − E[Xn ]
= E[(X − E[X]) (X − E[X])> ].

   
X1 Y1
X  Y 
 2  2
For two random vectors X =  .. , and Y =  .. , the (cross)-covariance
 .   . 
Xn Ym
matrix is given by
 
cov(X1 , Y1 ) cov(X1 , Y2 ) . . . cov(X1 , Ym )
cov(X, Y ) =  ..  ∈ Rn×m
.
cov(Xn , Y1 ) cov(Xn , Y2 ) . . . cov(Xn , Ym )
= E[(X − E[X]) (Y − E[Y ])> ].

26
 Sum of IID Random Variables

• In many applications, we need to repeat the experiment to generate more

samples. The outcome of every experiment is random and the experiments
are independent.

• Let Xi be the random variable that represents the outcome of i-th experi-
ment.

• The collection {Xi }i=1,2,...,N is said to be independent and identically dis-

tributed (IID) is each Xi has the same distribution and the random variables
in the collection are mutually independent.
Pn
• Suppose E[Xi ] = µ and var(Xi ) = σ 2 . Let S = i=1 Xi . Determine the
expectation and variance of S.

• Determine the characteristic function of S from the characteristic function

of Xi .

27
 Solution

Pn 1
Pn
Let S = i=1 X i and S̄ := n i=1 Xi .
 Pn  Pn
E[S] = E i=1 i =
X i=1 E[Xi ] = nµ.

E[S] = µ.

The variance of the sum is given by

n
X
Xi = E (S − E[S])2

 
var(S) = var
i=1
n
X
Xi − nµ)2
 
=E (
i=1
n
X
= E ( (Xi − µ))2

i=1
Xn X
(Xi − µ)2 +
 
=E (Xi − µ)(Xj − µ)
i=1 i6=j

= nσ 2 ,

since when two r.v.s Xi and Xj are independent,

 
E (Xi − E[Xi ])(Xj − E[Xj ]) = E[(Xi − E[Xi ])] × E[Xj − E[Xj ]] = 0.

σ2
Now: var(S̄) = ( n1 )2 var(S) = n

More generally, if var(X) = σ 2 , then var(cX) = c2 σ 2 .

28
 Gaussian Random Variable

A Gaussian random variable X is characterized by two parameters: mean (µ)

and variance σ 2 , and is denoted N (µ, σ 2 ). The distribution is defined below.
• If σ = 0, then P[X = µ] = 1 and P[X 6= µ] = 0.
• If σ > 0, it is a continuous random variable with density and CDF
Z c
1 − (x−µ)
2

fX (x) = √ e 2σ2 , Ψ(c) = fX (x)dx.

2πσ 2 −∞

Consequently, Z ∞
1 (x−µ)2
√ e− 2σ 2 = 1.
−∞ 2πσ 2

The characteristic function of X ∼ N (µ, σ 2 ) is given by

h2 σ 2
ΦX (h) = eiµh− 2 .

Most derivations involving Gaussian random variables and vectors leverage char-
acteristic function.

Suppose X1 , X2 , . . . , Xn be a collection ofPGaussian random variables and

are independent. Then, show that Z := ni=1 ai Xi is a Gaussian random
variable.

29
 Jointly Gaussian Random Variables

Definition 9. A collection of random variables (Xt )t∈T is called jointly

Gaussian if every finite linear combination is Gaussian. In particular,
X is a Gaussian random vector if its constituent random variables are
jointly Gaussian.
   
X1 Y1
X  Y 
 2  2
Two random vectors X =  ..  and Y =  ..  are jointly Gaussian if
 .   . 
Xn Ym
the collection {X1 , . . . , Xn , Y1 , . . . , Ym } is jointly Gaussian.

 
X1
X 
 2
A Gaussian random vector X =  ..  is characterized by two quantities:
 . 
Xn
 
E[X1 ]
 E[X ] 
2 
mean: µX = E[X] =  ..  ∈ Rn and

 . 
E[Xn ]
covariance matrix: CX ∈ Rn×n with (CX )i,j = cov(Xi , Xj ).

Show that the joint characteristic function of Gaussian random vector X is

given by
h> CX h
ih> µX −
ΦX (h) = e 2 .

30
 Properties of Gaussian Random Vectors

If X1 , X2 , . . . , Xn are jointly Gaussian, then each Xi is Gaussian.

If each of X1 , X2 , . . . , Xn are individually Gaussian and independent, then

the collection is jointly Gaussian.

If X is a Gaussian random vector, and Y = AX + b where A is a given

matrix and b is a given vector of suitable dimensions, then show that Y is a
Gaussian random vector, and find its mean and covariance.

If a collection of jointly Gaussian random variables are uncorrelated, then

they are independent.

Let X be a Gaussian random vector and V be another Gaussian random

vector uncorrelated with X. Let Y = AX + V where A is a given matrix.
Find the mean and covariance of Y . Is Y Gaussian? Does the answer change
when E[V ] = 0.

31
 Inequalities and Bounds

Pn
Union bound: If A1 , A2 ....... An are events, P(∪ni=1 Ai ) ≤ i=1 P(Ai )
(Equality holds when Ai are disjoint)

Markov’s Inequality: Let X be a non negative r.v. Then, for any  > 0,
E[X]
P(X ≥ ) ≤ .


1
Note: This bound is useful for large values of . In particular, if  < E[X] , then
E[X]
 > 1 which is trivial.

Main idea: Y ≤ X ⇒ E[Y ] ≤ E[X]. Define

(
, when X ≥ ,
Y =
0, otherwise.

Is Y ≤ X?, E[Y ] =?
Chebyshev’s Inequality: For any random variable X, with E[X] = µ, and
any  > 0,
var(X)
P[|X − µ| ≥ ] ≤ .
2

Proof: Apply Markov’s inequality to Y = (X − µ)2

var(S) σ2
Application: limn→∞ P[|S − µ| ≥ ] ≤ 2 = n2 = 0.

32
 Inequalities and Bounds

Hoeffding Inequality: Let X1P , X2 , .....Xn be independent random variables

with Xi ∈ [ai , bi ]. Let Sn = ni=1 Xi . Then,
2
  − PN 2
P Sn − E[Sn ] ≥  ≤ e i=1 i −ai )2 ,
(b

2
  − PN 2
P Sn − E[Sn ] ≤ − ≤ e i=1 (bi −ai )2 .

If Xi ’s are i.i.d. with ai = 0, bi = 1, then

 
Sn 2
P | − E[X1 ] |≥  ≤ 2e−2 n .
n

Discuss: Confidence interval using Hoeffding and Chebyshev.

Cauchy Schwartz Inequality: For two random variables X and Y ,

(E[XY ])2 ≤ E[X 2 ]E[Y 2 ]

Proof: Define Z = (sX + Y )2 ≥ 0. Then, for every s ∈ R,

E[Z] ≥ 0
=⇒ E[s2 X 2 + 2sXY + Y 2 ] ≥ 0
=⇒ s2 E[X 2 ] + 2sE[XY ] + E[Y 2 ] ≥ 0

Define: h(s) := s2 E[X 2 ] + 2sE[XY ] + E[Y 2 ]. Since h(s) ≥ 0 for all s ∈ R,

it does not have distinct real roots. From b2 − 4ac ≤ 0 formula for quadratic
functions, we obtain the inequality.

Corollary: Correlation coefficient lies in [−1, 1].

33
 Chernoff Bound

Note that P(X ≥ ) = P(etX ≥ et ) for any t > 0 since x ≥ y ⇔ ex ≥ ey .

From Markov’s inequality, we have

P(X ≥ ) = P(etX ≥ et )

E[etX ]
≤ for every t > 0
et
≤ min e−t E[etX ]

t>0
= min e−t mX (t)
 
t>0
= min elog(mX (t)−t)
 
t>0
 
− maxt>0 (t−log(mX (t))
=e ,

where E[etX ] = mX (t) is called the moment generating function of X.

Let X ∼ Binomial r.v (n,p) with probability mass function given by P(X =
k) = (nk)pk (1 − p)n−k , with k = {0, 1, 2.....n}. Find upper bounds on
P(X ≥ q) using Markov, Chebyshev and Chernoff bounds.

Homework: Plot the true probability and the bounds.

34
 Distribution of sum of two random variables

Let X1 and X2 be two continuous random variables,

Let us try to find distribution and density of X1 + X2 when
• X1 , X2 are arbitrary
• X1 , X2 are independent
• X1 , X2 are IID.

35
 Convergence of Sequences

Definition 10. A sequence of real numbers (xn )n∈N := (x1 , x2 , . . . , xn , . . .)

with each xi ∈ R, is said to converge to x∗ ∈ R if of every  > 0, there exists
n such that |xn − x∗ | <  for every x ≥ n . Then, we write limn→∞ xn = x∗ .

Note: The above definition requires us to first conjecture a limit point x∗ , which
may not always be trivial.

Definition 11. A sequence (xn )n∈N is a Cauchy sequence if

lim |xn − xm | = 0.
n,m→∞

Proposition: If a sequence (xn )n∈N is a Cauchy sequence, then it converges to

some finite limit.

Example: Let xn = n1 , i.e., the sequence (xn )n∈N = (1, 12 , 13 . . .). What is a
possible value of x∗ ? Is this sequence a Cauchy sequence?

Convergence of Random Variables:    

Consider a probability space (Ω, F, P) with Ω = [0, 1], and P [a, b] = P (a, b) =
b − a (uniform distribution). Let (Xn )n∈N is a sequence of random variables de-
fined on (Ω, F, P) with

Xn (ω) = ω n , ω ∈ [0, 1].

What do we mean by convergence of this sequence?

36
 Almost Sure Convergence

Definition 12 (Almost Sure Convergence). A sequence of random vari-

ables (Xn )n∈N converges almost surely to a random variable X ∗ if

P {ω : lim Xn (ω) = X ∗ (ω)} = 1.

 
n→∞

Equivalently, P {ω : limn→∞ Xn (ω) 6= X ∗ (ω)} = 0.

 

Note: For a given outcome ω, Xn (ω) is a sequence of real numbers.

Example: Does the sequence with Xn (ω) = ω n , ω ∈ [0, 1] converge almost

surely to some X ∗ ?

Example: Consider a sequence of random variables defined as:

X1 (ω) = 1, ω ∈ [0, 1],

X2 (ω) = 1, ω ∈ [0, 0.5],
X3 (ω) = 1, ω ∈ [0.5, 1],
X4 (ω) = 1, ω ∈ [0, 0.25],
X5 (ω) = 1, ω ∈ [0.25, 0.5],
X6 (ω) = 1, ω ∈ [0, 5, 0.75], and so on.

Does this sequence converge almost surely to some X ∗ ?

Let us determine the following quantities.

P(Xn 6= 0) =?
E[Xn ] =?
Both the above quantities define a sequence of real numbers. Do those sequence
converge?

37
 Convergence in Probability and in Mean Square Sense

Definition 13. A sequence of random variables (Xn )n∈N converges to a

r.v. X ∗ in probability, denoted Xn →P X ∗ , if

lim P[|Xn − X ∗ | ≥ ] = 0 for every  > 0.

n→∞

Definition 14. A sequence of random variables (Xn )n∈N , with E[Xn2 ] <
∞ ∀n, converges to X ∗ in mean square sense if

lim E (Xn − X ∗ )2 = 0.
 
n→∞

This is denoted by Xn →m.s X ∗ .

Example: Consider the following two sequence of random variables:

(
1, ω ∈ [0, n1 ]
Xn (ω) =
0, otherwise.
(
n, ω ∈ [0, n1 ]
Yn (ω) =
0, otherwise.

Determine if the above sequences converge almost surely, in probability and in

mean-square sense.

38
 Convergence in Distribution

Definition 15. A sequence of random variables (Xn )n∈N converges in

distribution to r.v. X ∗ if either of the following are true.
• limn→∞ FXn (x) = FX ∗ (x) at all points of continuity of FX ∗ .
∗
• limn→∞ E[eihXn ] = E[eihX ] for all h ∈ R.
• E[g(Xn )] → E[g(X ∗ )] for every bounded continuous function g.

Example
Let X be a r.v with CDF

α
 θ , α ∈ [0, θ],

FX (α) = 1, α ≥ θ,

0, α ≤ 0.


Let X1 , X2 , . . . , Xn be i.i.d with distribution FX . Define a sequence

Yk = maxi∈{1,2,...k} Xi .

Show that the sequence (Yn )n∈N converges in distribution to a random variable
Y ∗ whose distribution is given by
(
1, α ≥ θ
FY ∗ (α) =
0, otherwise.

• Convergence in probability implies convergence in distribution.

• Mean-square convergence implies convergence in probability.
• Almost sure convergence implies convergence in probability.

39
 Cauchy Criterion

Let (Xn )n∈N be a sequence of r.v defined on (Ω, F, P). Then,

• Xn converges almost surely to some random variable if

P[{ω : lim |Xm (ω) − Xn (ω)| = 0}] = 1.

m,n→∞

• Xn converges in probability to some r.v if

lim P[|Xm − Xn | > ] = 0.

m,n→∞

• Xn converges in m.s sense to some r.v if

lim E[(Xm − Xn )2 ] = 0.
n,m→∞

40
 Limit Theorems

Theorem 1 (Law of Large Numbers). Let X1 , X2 ..... be a sequence of

random
Pvariables. Each Xi has mean µX , i.e., E[Xi ] = µX . Define
n
Sn := i=1 Xi . Then,
Sn
• n →a.s
m.s µX if var(Xi ) ≤ C ∀i ∈ N and cov(Xi , Xj ) = 0 ∀i 6= j.
Sn
• If X1 , X2 , . . . i.i.d, then n →p µX (Weak law of large numbers)
Sn
• If X1 , X2 , . . . i.i.d, then n →a.s µX (Strong law of large numbers).

Sn
Note: What about the random variable n −µX ? What is its mean and variance?

Theorem 2 (Central Limit Theorem). Let each Xi be i.i.d, with E[Xi ] =

µX and var(Xi ) = σ 2 . Let N (µ, σ 2 ) denote Gaussian distribution with
mean µ and variance σ 2 . Then,
• Sn −µ
√ X n →d N (0, σ 2 ),


n
√ Sn
n −µX


• n σ →d N (0, 1),
√
• n Snn = Sn
√
n
→d N (µX , σ 2 ).

41